Gen AI models (stable diffusion)

[jtligon]

I would like a way to generate images locally using Stable Diffusion-esqe models.

ramalama serve --name stable-diffusion stable-diffusion

have it download a specified model, the allow me to ask the model for something and generate an image.

Would love to discuss the interface/tui/flow of this more with the team!

[ericcurtin]

https://diffusionbee.com/ an example of a similar tool

[ericcurtin]

This might be the best tool to look into to start as it uses the same library as llama.cpp:

https://github.com/leejet/stable-diffusion.cpp

GGML

[ericcurtin]

Here's a UI for it:

https://jellybox.com/

[rhatdan]

Is this something we care about or is llama.cpp need to get this functionality so we can take advantage?

[ericcurtin]

I don't expect llama.cpp to ever get this functionality, it's like whisper.cpp just another different form of AI inferencing

Although all three of these projects use ggml llama.cpp, whisper.cpp, stable-diffusion.cpp

[rhatdan]

Should stable-diffusion be considered a runtime then?

[jtligon]

Yes

[rhatdan]

PR to add stable-diffusion.cpp, one issue will be command line and how much disk space this adds. Could be added as a separate image if it is too big. Would be nice if it has same or similar CLI to llama-serve.

[rhatdan]

@jtligon interested in working on this?

[jtligon]

sure, I'll take it and start on it this week

[jtligon]

/assign

[github-actions[bot]]

A friendly reminder that this issue had no activity for 30 days.

[rhatdan]

@jtligon do you still consider this issue open?

[jtligon]

I'd say #1769 closed it. I wanted to do a cool /closed comment, but wasn't sure how.